The Use of Risk Prediction Models in Clinical Decision Making - PowerPoint PPT Presentation

1 / 60
About This Presentation
Title:

The Use of Risk Prediction Models in Clinical Decision Making

Description:

Peter Savage, M.D. National Heart, Lung, and Blood ... Michael A. Proschan, Ph.D. National Heart, Lung, and Blood Institute. J. Sanford Schwartz, M.D. ... – PowerPoint PPT presentation

Number of Views:214
Avg rating:3.0/5.0
Slides: 61
Provided by: sem6
Category:

less

Transcript and Presenter's Notes

Title: The Use of Risk Prediction Models in Clinical Decision Making


1
The Use of Risk Prediction Models in Clinical
Decision Making Public Health
Surveillance.
  • Christopher T. Sempos, Ph.D.
  • Department of Social and Preventive Medicine
  • University at Buffalo, Buffalo, NY, USA

2
Outline
  • Overview
  • Risk Prediction Models used in the treatment of
    hyperlipidemia
  • WNY Cardiovascular Risk Reduction Initiative
  • Evaluation of Framingham risk models
  • Unresolved Issues

3
The Use of Risk Prediction Models in Clinical
Decision Making
  • Overview

4
Two key steps in rationale therapeutic decision
making in hypercholesterolemia
  • Clinical trial data.
  • Individual benefit of treatment should be based
    on pretreatment absolute coronary risk.

Simon A, et al. Circulation 1997962449-2452
5
Clinical Uses of Absolute Risk Prediction
  • Risk charts and tables
  • Pen and pencil score sheets
  • Slide rules
  • Hand held calculators
  • Web based
  • Palm Pilots

6
Functions of Risk Factor Charts
  • Predict a patients risk of disease during a
    specified time range.
  • Point out to younger patients how risk will
    increase with age even without an increase in
    risk factor levels.
  • Estimate relative risk.
  • Estimate change in risk with changes in risk
    factor levels

7
Proportion of CHD Cases Predicted by Established
Risk Factors
  • Blood Cholesterol, high blood pressure
    cigarette smoking explain at least 75 of CHD
    cases.
  • Add diabetes to list and explain about 85 or
    more of CHD cases.

Source Magnus P Beaglehole R. Arch Intern Med
20011612657-60.
8
Risk Prediction Models used in the treatment of
hyperlipidemia
  • US, Australian and European Guidelines

9
US, Australian/NZ European Lipid Guidelines
  • US National Cholesterol Education Programs
    Third Adult Treatment Panel Guidelines (ATP III)
    May, 2001.
  • Australia/NZ National Heart Foundation of
    Australia November, 2001.
  • Europe Recommendations of the Second Joint Task
    Force of European and Other Societies on Coronary
    Prevention - 1998.

10
Third Report of the Expert Panel on Detection,
Evaluation, and Treatment of High Blood
Cholesterol in Adults (Adult Treatment Panel
III) (ATP III)
Members Scott M. Grundy, M.D., Ph.D.
(Chair) University of Texas Southwestern Medical
Center at Dallas Diane Becker, Sc.D., M.P.H.
The Johns Hopkins University Luther T. Clark,
M.D. State University of New York,
Brooklyn Richard S. Cooper, M.D. Loyola
University Medical School Margo A. Denke,
M.D. University of Texas Southwestern Medical
Center at Dallas Wm. James Howard,
M.D. Washington Hospital Center Donald B.
Hunninghake, M.D. University of Minnesota
D. Roger Illingworth, M.D., Ph.D. The Oregon
Health Sciences University Russell V. Luepker,
M.D., M.S. University of Minnesota Patrick
McBride, M.D., M.P.H. University of Wisconsin
Hospital and Clinics James M. McKenney,
Pharm.D. National Clinical Research Richard C.
Pasternak, M.D., F.A.C.C. Massachusetts General
Hospital Neil J. Stone, M.D. Northwestern
University School of Medicine Linda Van Horn,
Ph.D, R.D. Northwestern University Medical
School
Ex-Officio Members H. Bryan Brewer, Jr.,
M.D. National Heart, Lung, and Blood
Institute James I. Cleeman, M.D. (Executive
Director) National Heart, Lung, and Blood
Institute Nancy D. Ernst, Ph.D., R.D. National
Heart, Lung, and Blood Institute David Gordon,
M.D., Ph.D. National Heart, Lung, and Blood
Institute Daniel Levy, M.D. National Heart,
Lung, and Blood Institute Basil Rifkind,
M.D. National Heart, Lung, and Blood Institute
Jacques E. Rossouw, M.D. National Heart, Lung,
and Blood Institute Peter Savage, M.D. National
Heart, Lung, and Blood Institute Consultants Ste
ven M. Haffner, M.D. University of Texas Health
Science Center, San Antonio David G. Orloff,
M.D. Food and Drug Administration Michael A.
Proschan, Ph.D. National Heart, Lung, and Blood
Institute J. Sanford Schwartz, M.D. University
of Pennsylvania Christopher T. Sempos, PhD State
University of New York, Buffalo
11
Major Changes in ATP III
  • New Risk Assessment Approach with an increased
    emphasis on determining absolute CHD risk.
  • LDL
  • Low HDL defined as
  • Lower triglyceride cut-points (
  • Intensified emphasis on lifestyle change.

12
Risk Categories for Patient Detection and
Evaluation in ATP III Guidelines
  • CHD or CHD Risk Equivalent
  • CHD or other atherosclerotic diseases
  • Diabetes
  • 10 Year Predicted Risk of Hard CHD 20
  • Multiple (2) Risk Factors and 10 Year Predicted
    Risk 10-20
  • Multiple (2) Risk Factors and 10 Year Predicted
    Risk
  • Zero to 1 Risk Factor

13
LDL Cholesterol Goals Cutpoints (mg/dL) ATP
III Guidelines
May be revised to 100 mg/dL After 3 months
trial of diet lifestyle.
14
Use of Risk Prediction Models in International
Guidelines
  • US Risk factor counting and three levels of the
    10-year hard CHD risk using a Framingham model
    ( 20, 10-20
  • Australia 5-year CVD risk 10-15 using a
    Framingham model or risk factor counting.
  • Europe 10-year total CHD risk 20 now or as
    projected to age 60 using a Framingham model.

15
Source of 20 Risk in 10 Years
  • Jackson R, et al. BMJ 1993307107-110.
  • Authors suggested
  • Using absolute risk for treatment of hypertension
  • Only those with 20 risk of CVD in 10 years
    would be treated
  • Absolute risk based on a Framingham risk score.

16
Risk Factors Used In ATP IIIFramingham Risk
Models
  • Age
  • Systolic Blood Pressure (mm Hg)
  • Serum Total Cholesterol (mg/dL)
  • HDL Cholesterol (mg/dL)
  • Cigarette smoking history
  • Use of Antihypertensive medications
  • (Only if SBP 120 mm Hg)

17
Risk Prediction Models used in public health
surveillance
  • The WNY Cardiovascular Risk Reduction Initiative

18
The WNY Cardiovascular Risk Reduction Initiative
  • Designed to reduce predicted CVD risk by 25 over
    the next 10 years.
  • CVD risk reduction in 3 main HMOs
  • Reduction in risk to be accomplished by
  • ? Patient knowledge cholesterol BP levels
  • ? Proportion of in treatment by guidelines
  • ? Proportion of patients treated to goal

19
The WNY Cardiovascular Risk Reduction Initiative
  • Initiative will consist of a series of 5
    cross-sectional studies conducted every other
    year.
  • Sample size of 5,000 plan participants
  • Two types of data collected
  • Patient CVD Risk Survey
  • Chart Review

20
The WNY Cardiovascular Risk Reduction Initiative
  • Patient CVD Risk Survey
  • Have lipids BP been checked
  • High?
  • Chart Review
  • ATP III JNC VI clinical information

21
Ten Year Framingham Predicted Risk
ATP III Framingham Risk Model. Personal
Communication Dr. Lisa Sullivan Anderson K, et
al. Am Heart J. 1991121293-8.
22
Sample Sizes 25 Reduction in Predicted Risk
Alpha 0.05, Power 80. H ATP III Framingham
Risk Model. Personal Communication Dr. Lisa
Sullivan. Anderson K, et al. Am Heart J.
1991121293-8.
23
Variables in Australian/NZ EuropeanFramingham
Model
  • Age
  • Sex
  • Total Cholesterol
  • HDL Cholesterol
  • Smoking
  • Blood Pressure
  • Diabetes
  • ECG-LVH

Modeled as Total/HDL ratio
24
The Use of Risk Prediction Models in Clinical
Decision Making
  • Evaluation of Framingham risk models

25
Framingham, Massachusetts
26
Why Framingham?
  • Long-term follow-up
  • Cohort re-examined on a regular schedule
  • CVD incidence and mortality
  • Consistent case ascertainment
  • Estimates of risk for men and women

27
Criteria Used to Evaluate Predictive Accuracy of
Framingham Risk Functions to Other Populations
  • Ordering Risk in individuals
  • Estimates of Relative Risk
  • Prediction of absolute risk

28
The Multivariate Risk of CVD in Diverse
Populations Project
  • 16 Observational Studies
  • Examine issue in multivariate risk
  • 161,955 persons (men 105,420 women 56,535)
  • 29,662 deaths (men 21,841 women 7,821)
  • 8,316 CHD deaths (men 6,442 women 1,874)
  • Follow-up time ranges from 8-20 years
  • Dr. Daniel McGee - PI

29
Diverse Populations Collaborative Group Studies
  • Framingham
  • NHANES I
  • NHANES II
  • Tecumseh
  • Honolulu Heart Prog.
  • LRC Prevalence Study
  • Yugoslavia
  • Scottish Coll. Study
  • Puerto Rico HH Study
  • Renfrew and Paisley
  • The Glostrup Cohort
  • Norwegian Counties
  • Israeli Ischemic H.D.
  • LRC-CPPT
  • HDFP
  • MRFIT

30
Baseline variables used to predict risk of
CHD death
  • Age (years)
  • Systolic blood pressure (mm/Hg)
  • Serum total cholesterol (mg/dL)
  • Current cigarette smoking status (yes/no)
  • Current diabetes status (yes/no)

31
Criteria to Evaluate Risk Function Accuracy
  • Ordering Risk in individuals
  • Areas under Receiver Operator Curves (AUC)

32
Areas under the receiver operating characteristic
(AUC) curves of 8-year death from
coronary heart disease.
33
Areas under the receiver operating characteristic
(AUC) curves of 8-year death from coronary
heart disease.
34
Criteria to Evaluate Risk Function Accuracy
Estimates of Relative Risk (RR)
  • Risk factors significantly related to CHD.
  • RR estimates from regression coefficients
  • Magnitude of RR estimates varied.
  • Significant heterogeneity for all variables
    except age.

35
Criteria to Evaluate Risk Function Accuracy
  • Prediction of absolute risk

36
Predicted observed 8-year risk of CHD death
using a Framingham Model
37
Criteria to Evaluate Risk Function Accuracy
Summary
  • AUCs similar using different risk functions from
    Framingham cohort specific models
  • Ability to rank related to intrinsic
    discriminatory power of risk factors not
    multivariate model.
  • RR estimates varied significantly.
  • Prediction of absolute risk not accurate.

38
Appropriateness of Framingham models to other
populations
  • Ranks persons similarly to study specific models
  • Discriminates between cases and non-cases of CHD
    similarly to study specific models
  • Absolute risk?

See DAgostino et al. JAMA 2001286180-187.
39
Predictive Ability of Framingham Risk Models
Original Cohort
Adjusted for Age, LDL, hypertension, diabetes,
smoking, BMI, SBP HDL
40
Predictive Ability of Framingham Risk Models
Offspring Cohort
Adjusted for Age, LDL, hypertension, diabetes,
smoking, BMI, SBP HDL
41
Statistical Models Used in Framingham Risk
Prediction
  • Logistic Model
  • Probability (p) 1/(1 e-? ?ixi)
  • Cox proportional Hazards Model
  • Probability (p) 1 S(t)exp(?x - ??)
  • Accelerated Failure Time (Weibull Model)

ATP III risk prediction based on a Cox model.
42
Options for Modifying Framingham Risk Model Cox
Model
  • Replace Framingham Beta Coefficients with study
    specific values.
  • Replace Framingham Mean Values with Study
    Specific Means.
  • Replace Average Survival rate (S(t)) with study
    specific value.

43
The Use of Risk Prediction Models in Clinical
Decision Making
  • Unresolved Issues

44
Issues in Absolute Risk PredictionUsing
Framingham Models
  • Appropriate CVD endpoint?
  • No trial data based on multivariate risk
  • Appropriate time frame to measure risk - 5 yr.,
    10 yr., lifetime?
  • Appropriate level or risk 20?
  • What variables how many variables to include?
  • Confidence limits on risk estimate?
  • Appropriateness of Framingham models to other
    populations?

45
Thank you!
  • Questions or Comments?

46
Selected References
  • Gordon T, Kannel W. Am. Heart J. June
    19821031-1039.
  • Simon A, et al. Circulation 1997962449-52.
  • Grundy et al. Circulation 19991001481-92.
  • DAgostino et al. JAMA 2001286180-87.
  • Diverse Populations Collaborative Group Heart
    200288222-228.

47
ATP III Risk Factors
  • Positive Risk Factors
  • Cigarette smoking
  • Hypertension (BP140/90 mmHg or meds)
  • Low HDL (
  • Family History of Premature CHD
  • Age men 45 women 55 years
  • Negative Risk Factor
  • HDL cholesterol 60 mg/dL

48
Drug Therapy PercentFramingham Models
TC 160 mg/dL TC 190 mg/dL
49
Drug Therapy PercentFramingham Models
50
ATP III Framingham Models
51
ATP III Framingham Models
52
(No Transcript)
53
European Guidelines
54
Counting Risk Factors Australian vs. US
Guidelines
  • Use of overweight/obesity as a risk factor
  • Lower age threshold for women 45 vs. 55
  • High HDL not used as a negative risk factor
  • Impaired fasting glucose/microalbuminuria
  • Higher lipid threshold for primary prevention
    LDL - 4.0) mmol/L vs various levels (3.4, 4.1
    5.0 mmol/L).

55
Framingham Models Australian vs. European
Guidelines
  • CVD vs. CHD
  • Direct use of drug therapy in persons with CHD,
    i.e. no period of diet therapy.
  • Diabetes defined as a CHD risk equivalent.
  • Lower lipid threshold for drug therapy
    Total-C 4.0 vs. 5.0 mmol/L

56
Variables in Framingham Models
57
Variables in Framingham Models
58
Guidelines - USA
  • The Executive Summary of the Third Report of the
    National Cholesterol Education Program (NCEP)
    Expert Panel on Detection, Evaluation, and
    Treatment of High Blood Cholesterol in Adults
    (Adult Treatment Panel III or ATP III).
  • JAMA 2001(May 16)2852486-2497.

59
Guidelines - Australia
  • Lipid Management Guidelines - 2001
  • The Medical Journal of Australia 2001175 (5
    Nov. Supplement)S55-S88.
  • Principal Sponsoring Organizations National
    Heart Foundation of Australia (NHFA) and the
    Cardiac Society of Australia and New Zealand.

60
Guidelines - Europe
  • Prevention of Coronary Heart Disease in Clinical
    Practice Recommendations of the Second Joint
    Task Force of European and Other Societies on
    Coronary Prevention.
  • Wood D et al. European Heart Journal
    1998191434-1502.
Write a Comment
User Comments (0)
About PowerShow.com